The line you're quoting comes from a RuntimeWire write-up dated October 25, 2026. It reports that Scott Armstrong, a PDE mathematician (NYU Courant Institute and Sorbonne Université), called OpenAI's Navier-Stokes construction 'unbelievable' and said it may be beyond what a human could have devised, though not beyond human understanding. RuntimeWire That wording closely echoes Greg Brockman's praise from the same week, so the article may be framing Armstrong's reaction against Brockman's. I could not open Armstrong's full post, so I can't confirm his exact reasoning.
His own posts in search results are more critical on the technical side. He cites a new paper by Constantin, Ignatova and Vicol arguing that OpenAI's blow-up construction cannot produce an analytic forcing term, so it does not work with zero forcing, the formulation (A) of the Clay problem. Armstrong on X
For context, OpenAI's claim (September 2026) is a finite-time singularity for the forced Navier-Stokes equations, with a Lean formalization. OpenAI Critics note that the forced version is the one the Clay statement allows, but not the harder unforced case. Science Whether the proof and its Lean formalization fully match is still being disputed, so treat the claim as unsettled.
🕒 As of October 2026
